1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Common helpers for stackable filesystems and backing files.
4 *
5 * Forked from fs/overlayfs/file.c.
6 *
7 * Copyright (C) 2017 Red Hat, Inc.
8 * Copyright (C) 2023 CTERA Networks.
9 */
10
11#include <linux/fs.h>
12#include <linux/backing-file.h>
13#include <linux/splice.h>
14#include <linux/mm.h>
15
16#include "internal.h"
17
18/**
19 * backing_file_open - open a backing file for kernel internal use
20 * @user_path: path that the user reuqested to open
21 * @flags: open flags
22 * @real_path: path of the backing file
23 * @cred: credentials for open
24 *
25 * Open a backing file for a stackable filesystem (e.g., overlayfs).
26 * @user_path may be on the stackable filesystem and @real_path on the
27 * underlying filesystem. In this case, we want to be able to return the
28 * @user_path of the stackable filesystem. This is done by embedding the
29 * returned file into a container structure that also stores the stacked
30 * file's path, which can be retrieved using backing_file_user_path().
31 */
32struct file *backing_file_open(const struct path *user_path, int flags,
33 const struct path *real_path,
34 const struct cred *cred)
35{
36 struct file *f;
37 int error;
38
39 f = alloc_empty_backing_file(flags, cred);
40 if (IS_ERR(ptr: f))
41 return f;
42
43 path_get(user_path);
44 backing_file_set_user_path(f, path: user_path);
45 error = vfs_open(real_path, f);
46 if (error) {
47 fput(f);
48 f = ERR_PTR(error);
49 }
50
51 return f;
52}
53EXPORT_SYMBOL_GPL(backing_file_open);
54
55struct file *backing_tmpfile_open(const struct path *user_path, int flags,
56 const struct path *real_parentpath,
57 umode_t mode, const struct cred *cred)
58{
59 struct mnt_idmap *real_idmap = mnt_idmap(mnt: real_parentpath->mnt);
60 struct file *f;
61 int error;
62
63 f = alloc_empty_backing_file(flags, cred);
64 if (IS_ERR(ptr: f))
65 return f;
66
67 path_get(user_path);
68 backing_file_set_user_path(f, path: user_path);
69 error = vfs_tmpfile(idmap: real_idmap, parentpath: real_parentpath, file: f, mode);
70 if (error) {
71 fput(f);
72 f = ERR_PTR(error);
73 }
74 return f;
75}
76EXPORT_SYMBOL(backing_tmpfile_open);
77
78struct backing_aio {
79 struct kiocb iocb;
80 refcount_t ref;
81 struct kiocb *orig_iocb;
82 /* used for aio completion */
83 void (*end_write)(struct kiocb *iocb, ssize_t);
84 struct work_struct work;
85 long res;
86};
87
88static struct kmem_cache *backing_aio_cachep;
89
90#define BACKING_IOCB_MASK \
91 (IOCB_NOWAIT | IOCB_HIPRI | IOCB_DSYNC | IOCB_SYNC | IOCB_APPEND)
92
93static rwf_t iocb_to_rw_flags(int flags)
94{
95 return (__force rwf_t)(flags & BACKING_IOCB_MASK);
96}
97
98static void backing_aio_put(struct backing_aio *aio)
99{
100 if (refcount_dec_and_test(r: &aio->ref)) {
101 fput(aio->iocb.ki_filp);
102 kmem_cache_free(s: backing_aio_cachep, objp: aio);
103 }
104}
105
106static void backing_aio_cleanup(struct backing_aio *aio, long res)
107{
108 struct kiocb *iocb = &aio->iocb;
109 struct kiocb *orig_iocb = aio->orig_iocb;
110
111 orig_iocb->ki_pos = iocb->ki_pos;
112 if (aio->end_write)
113 aio->end_write(orig_iocb, res);
114
115 backing_aio_put(aio);
116}
117
118static void backing_aio_rw_complete(struct kiocb *iocb, long res)
119{
120 struct backing_aio *aio = container_of(iocb, struct backing_aio, iocb);
121 struct kiocb *orig_iocb = aio->orig_iocb;
122
123 if (iocb->ki_flags & IOCB_WRITE)
124 kiocb_end_write(iocb);
125
126 backing_aio_cleanup(aio, res);
127 orig_iocb->ki_complete(orig_iocb, res);
128}
129
130static void backing_aio_complete_work(struct work_struct *work)
131{
132 struct backing_aio *aio = container_of(work, struct backing_aio, work);
133
134 backing_aio_rw_complete(iocb: &aio->iocb, res: aio->res);
135}
136
137static void backing_aio_queue_completion(struct kiocb *iocb, long res)
138{
139 struct backing_aio *aio = container_of(iocb, struct backing_aio, iocb);
140
141 /*
142 * Punt to a work queue to serialize updates of mtime/size.
143 */
144 aio->res = res;
145 INIT_WORK(&aio->work, backing_aio_complete_work);
146 queue_work(wq: file_inode(f: aio->orig_iocb->ki_filp)->i_sb->s_dio_done_wq,
147 work: &aio->work);
148}
149
150static int backing_aio_init_wq(struct kiocb *iocb)
151{
152 struct super_block *sb = file_inode(f: iocb->ki_filp)->i_sb;
153
154 if (sb->s_dio_done_wq)
155 return 0;
156
157 return sb_init_dio_done_wq(sb);
158}
159
160static int do_backing_file_read_iter(struct file *file, struct iov_iter *iter,
161 struct kiocb *iocb, int flags)
162{
163 struct backing_aio *aio = NULL;
164 int ret;
165
166 if (is_sync_kiocb(kiocb: iocb)) {
167 rwf_t rwf = iocb_to_rw_flags(flags);
168
169 return vfs_iter_read(file, iter, ppos: &iocb->ki_pos, flags: rwf);
170 }
171
172 aio = kmem_cache_zalloc(backing_aio_cachep, GFP_KERNEL);
173 if (!aio)
174 return -ENOMEM;
175
176 aio->orig_iocb = iocb;
177 kiocb_clone(kiocb: &aio->iocb, kiocb_src: iocb, filp: get_file(f: file));
178 aio->iocb.ki_complete = backing_aio_rw_complete;
179 refcount_set(r: &aio->ref, n: 2);
180 ret = vfs_iocb_iter_read(file, iocb: &aio->iocb, iter);
181 backing_aio_put(aio);
182 if (ret != -EIOCBQUEUED)
183 backing_aio_cleanup(aio, res: ret);
184 return ret;
185}
186
187ssize_t backing_file_read_iter(struct file *file, struct iov_iter *iter,
188 struct kiocb *iocb, int flags,
189 struct backing_file_ctx *ctx)
190{
191 ssize_t ret;
192
193 if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
194 return -EIO;
195
196 if (!iov_iter_count(i: iter))
197 return 0;
198
199 if (iocb->ki_flags & IOCB_DIRECT &&
200 !(file->f_mode & FMODE_CAN_ODIRECT))
201 return -EINVAL;
202
203 scoped_with_creds(ctx->cred)
204 ret = do_backing_file_read_iter(file, iter, iocb, flags);
205
206 if (ctx->accessed)
207 ctx->accessed(iocb->ki_filp);
208
209 return ret;
210}
211EXPORT_SYMBOL_GPL(backing_file_read_iter);
212
213static int do_backing_file_write_iter(struct file *file, struct iov_iter *iter,
214 struct kiocb *iocb, int flags,
215 void (*end_write)(struct kiocb *, ssize_t))
216{
217 struct backing_aio *aio;
218 int ret;
219
220 if (is_sync_kiocb(kiocb: iocb)) {
221 rwf_t rwf = iocb_to_rw_flags(flags);
222
223 ret = vfs_iter_write(file, iter, ppos: &iocb->ki_pos, flags: rwf);
224 if (end_write)
225 end_write(iocb, ret);
226 return ret;
227 }
228
229 ret = backing_aio_init_wq(iocb);
230 if (ret)
231 return ret;
232
233 aio = kmem_cache_zalloc(backing_aio_cachep, GFP_KERNEL);
234 if (!aio)
235 return -ENOMEM;
236
237 aio->orig_iocb = iocb;
238 aio->end_write = end_write;
239 kiocb_clone(kiocb: &aio->iocb, kiocb_src: iocb, filp: get_file(f: file));
240 aio->iocb.ki_flags = flags;
241 aio->iocb.ki_complete = backing_aio_queue_completion;
242 refcount_set(r: &aio->ref, n: 2);
243 ret = vfs_iocb_iter_write(file, iocb: &aio->iocb, iter);
244 backing_aio_put(aio);
245 if (ret != -EIOCBQUEUED)
246 backing_aio_cleanup(aio, res: ret);
247 return ret;
248}
249
250ssize_t backing_file_write_iter(struct file *file, struct iov_iter *iter,
251 struct kiocb *iocb, int flags,
252 struct backing_file_ctx *ctx)
253{
254 ssize_t ret;
255
256 if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
257 return -EIO;
258
259 if (!iov_iter_count(i: iter))
260 return 0;
261
262 ret = file_remove_privs(iocb->ki_filp);
263 if (ret)
264 return ret;
265
266 if (iocb->ki_flags & IOCB_DIRECT &&
267 !(file->f_mode & FMODE_CAN_ODIRECT))
268 return -EINVAL;
269
270 scoped_with_creds(ctx->cred)
271 return do_backing_file_write_iter(file, iter, iocb, flags, end_write: ctx->end_write);
272}
273EXPORT_SYMBOL_GPL(backing_file_write_iter);
274
275ssize_t backing_file_splice_read(struct file *in, struct kiocb *iocb,
276 struct pipe_inode_info *pipe, size_t len,
277 unsigned int flags,
278 struct backing_file_ctx *ctx)
279{
280 ssize_t ret;
281
282 if (WARN_ON_ONCE(!(in->f_mode & FMODE_BACKING)))
283 return -EIO;
284
285 scoped_with_creds(ctx->cred)
286 ret = vfs_splice_read(in, ppos: &iocb->ki_pos, pipe, len, flags);
287
288 if (ctx->accessed)
289 ctx->accessed(iocb->ki_filp);
290
291 return ret;
292}
293EXPORT_SYMBOL_GPL(backing_file_splice_read);
294
295ssize_t backing_file_splice_write(struct pipe_inode_info *pipe,
296 struct file *out, struct kiocb *iocb,
297 size_t len, unsigned int flags,
298 struct backing_file_ctx *ctx)
299{
300 ssize_t ret;
301
302 if (WARN_ON_ONCE(!(out->f_mode & FMODE_BACKING)))
303 return -EIO;
304
305 if (!out->f_op->splice_write)
306 return -EINVAL;
307
308 ret = file_remove_privs(iocb->ki_filp);
309 if (ret)
310 return ret;
311
312 scoped_with_creds(ctx->cred) {
313 file_start_write(file: out);
314 ret = out->f_op->splice_write(pipe, out, &iocb->ki_pos, len, flags);
315 file_end_write(file: out);
316 }
317
318 if (ctx->end_write)
319 ctx->end_write(iocb, ret);
320
321 return ret;
322}
323EXPORT_SYMBOL_GPL(backing_file_splice_write);
324
325int backing_file_mmap(struct file *file, struct vm_area_struct *vma,
326 struct backing_file_ctx *ctx)
327{
328 struct file *user_file = vma->vm_file;
329 int ret;
330
331 if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
332 return -EIO;
333
334 if (!can_mmap_file(file))
335 return -ENODEV;
336
337 vma_set_file(vma, file);
338
339 scoped_with_creds(ctx->cred)
340 ret = vfs_mmap(file: vma->vm_file, vma);
341
342 if (ctx->accessed)
343 ctx->accessed(user_file);
344
345 return ret;
346}
347EXPORT_SYMBOL_GPL(backing_file_mmap);
348
349static int __init backing_aio_init(void)
350{
351 backing_aio_cachep = KMEM_CACHE(backing_aio, SLAB_HWCACHE_ALIGN);
352 if (!backing_aio_cachep)
353 return -ENOMEM;
354
355 return 0;
356}
357fs_initcall(backing_aio_init);
358

source code of linux/fs/backing-file.c